S5 Optimization in context
Optimizing high-tech systems towards their operational context aims at ensuring the best possible match between a system’s setup and its environment, task at hand, and the surrounding system-of-systems. Improving effectiveness and efficiency, it ensures lasting performance and continued fitness even within diverse and highly dynamic situations and circumstances.
In this series of talks we illustrate different optimization scenarios, ranging from a system’s self-calibration towards adaptations within highly dynamic environments, and show how system-models within Digital Twins and even AI-based machine-reasoning and learning can be used to address them.
The optimal operation of high-tech equipment like ASML’s lithography technology allows us to drive our systems at the edge of the physical possible. This is fundamental to many industrial processes, like for mass producing semiconductor chips. It is also a fascinating challenge on various levels, as it bonds sub-system calibration with machine software parametrization, tunes into a multi-vendor system-of-systems setup and responds to the goals and settings of automation systems that steer multi-billion $ semiconductor fabrication plants.
Solving this challenge and understanding how one’s systems contributes best within a larger context opens the door towards smart industry, where AI and machine learning can offer ever more powerful, faster, and energy efficient solutions, providing they meet our highest standards – which is where research and engineering must meet.
Willeke van Vught, TNO
Adaptive system behavior for highly dynamic situations
Unmanned mobile systems generally need frequent operator interventions to operate effectively and safely in complex and unpredictable environments. As the complexity of the environment increases, the number of operator interventions increases substantially. To tackle this problem, we develop hybrid AI capabilities for real time self-assessment of a system's capabilities, its performance in context, and possible interventions that optimize the system's effectiveness. This allows systems to self-optimize in highly dynamic environments, achieving lasting fitness and a higher degree of autonomy, thus increasing their effectiveness while reducing the number of operator interventions.
Maurits Diephuis, ThermoFisher Scientific
and Ilona Armengol Thijs, ESI (TNO)
Learning in Digital Twins to automated the calibration of high-tech systems
In electron microscopy, the calibration of diverse subsystems is key to acquire a high quality image. Even though some assisting tooling exists, such a calibration process is usually lengthy and requires a highly skilled operator.
Automating the calibration processes, either in parts or fully, is therefore desirable to ease the burden on these experts and to reduce total-costs-of-ownership, as well as to improve the quality of the resulting image.
In this talk, we describe our achievements in electron microscopy calibration automation using reinforcement learning and Digital Twins. Reinforcement learning, an AI training method that rewards desired behaviors and/or punishing undesired ones, allows us here to automate decision taking in sequential processes – such as calibration, but also, more general, any optimization of a high-tech system or sub-system towards its individual context.